A convolutional neural network based approach to financial time series prediction

Area of Science:

• Quantitative Finance
• Computational Economics
• Applied Mathematics

Background:

• Financial time series exhibit chaotic behavior, making prediction complex.
• Existing prediction models often struggle with the inherent non-linearity and unpredictability of financial data.

Purpose of the Study:

• To develop and validate a novel hybrid model for financial time series prediction.
• To enhance prediction accuracy by integrating Chaos Theory with advanced machine learning techniques.
• To assess the hybrid model's performance against established forecasting methods.

Main Methods:

• A hybrid model integrating Chaos Theory, Convolutional Neural Networks (CNN), and Polynomial Regression (PR) was developed.
• Chaos Theory was employed to model the chaotic dynamics within financial time series.
• CNN provided initial predictions, and PR refined these predictions by modeling the error series.
• The hybrid model (Chaos+CNN+PR) was evaluated using diverse financial datasets: foreign exchange rates, commodity prices, and stock market indices.

Main Results:

• The proposed Chaos+CNN+PR hybrid model significantly outperformed benchmark models including ARIMA, Prophet, CART, RF, CNN, Chaos+CART, Chaos+RF, and Chaos+CNN.
• Superiority was demonstrated across multiple error metrics: Mean Squared Error (MSE), Mean Absolute Percentage Error (MAPE), Dstat, and Theil's U.
• The model showed robust performance across various financial time series types.

Conclusions:

• The hybrid Chaos+CNN+PR model offers a powerful and effective approach for financial time series prediction.
• Integrating chaos modeling with deep learning and regression techniques enhances predictive accuracy.
• This hybrid methodology represents a significant advancement in quantitative finance and algorithmic trading.